home›LLMs›

xAI Completes Grok V9-Medium Training, June Release Expected

Elon Musk confirms 1.5T parameter model, tripling its predecessor, now enters fine-tuning for a public launch in weeks with enhanced coding capabilities.

May 26, 2026

#Dev Tools #Fine Tuning #LLM #Reinforcement Learning #Training

xAI has finished training its Grok V9-Medium foundational model, a 1.5 trillion parameter AI with significant improvements over its predecessor, v8-small. The model, which heavily emphasizes coding tasks through Cursor data, is now undergoing fine-tuning and reinforcement learning, with a public release anticipated in early to mid-June 2026.

Grok V9-Medium Training Completed, Release Expected in June

On Sunday, May 24, 2026, Elon Musk confirmed that xAI had finished training the Grok V9-Medium foundational model. Evaluations of the completed base model are positive, according to reports the following day. The model now enters the fine-tuning and reinforcement learning stages ahead of a public launch within two to three weeks.

Model Scale Triples Over Predecessor

Grok V9-Medium is a 1.5 trillion parameter model, a threefold increase from the current production system. For comparison, the model serving all Grok traffic today is v8-small, which runs on approximately 0.5 trillion parameters.

Musk previously acknowledged that v8-small had issues:

Training data quality was insufficient
Data comprehensiveness and balance fell short

The jump to 1.5T parameters with a new training methodology aims to address those gaps.

A luminous crystalline sphere suspended in infinite darkness, its surface alive with cascading fractal patterns that branch outward in triplicate, each new tendril spawning two further extensions until the original form dwarfs into a single bright seed surrounded by an expansive constellation of interconnected geometric nodes, their surfaces rippling with bioluminescent pulses like synaptic firing, light bleeding from within the structures in waves of blue-white radiance that dissolve into the void, the overall composition conveying overwhelming scale through perspective compression as if witnessing something minute transform into something astronomical, atmosphere of contained power and exponential growth, cinematic lighting, hyperdetailed rendering, octane render, volumetric god rays, 8k resolution

From Base Training to Public Release

With the base model complete, xAI has moved Grok V9-Medium into supervised fine-tuning (SFT). Reinforcement learning (RL) training is scheduled to start within days. The company confirmed these phases are on track.

The public release is expected in early to mid-June 2026. The timeline hinges on smooth execution of the remaining tuning stages.

Heavy Emphasis on Coding via Cursor Data

A notable part of the training pipeline is the ingestion of large amounts of Cursor code data. Cursor is an AI-powered coding assistant widely used by developers.

Musk stated that Grok V9-Medium significantly outperforms v8-small on complex programming tasks. The model is shaped to handle:

Code generation
Debugging
Instruction-following in software development contexts

Additional Cursor data is planned for future training cycles, signaling a sustained focus on developer workflows.

Hardware and Benchmark Silence

Grok V9-Medium has been optimized for NVIDIA Blackwell architecture GPUs. xAI has not released any benchmark numbers for the model at this stage. The optimization indicates a partnership or alignment with the latest NVIDIA hardware for inference and training efficiency.

Strategic Positioning and Broader Context

xAI continues to expand its AI infrastructure, with Grok primarily integrated into the X platform. The coding-oriented capabilities of V9-Medium suggest a future as a developer-focused tool.

Possible downstream integrations include:

Tesla vehicles
The X platform itself, enhancing in-app developer assistance

The model’s design reflects a deliberate shift toward practical, task-specific performance, particularly in coding, rather than general-purpose benchmarks.